Cerebras Wafer-Scale cluster (R2.2.1)

Get started

Get started by setting up your Cerebras virtual environment. Try out our state-of-the-art deep learning models that reside in our Model Zoo and proceed to launch your first job in the Cerebras Wafer-Scale cluster:

• Set up a Cerebras virtual environment
• Clone the Cerebras Model Zoo
• Launch your first job on the Cerebras Wafer-Scale cluster

Familiarize yourself with the Cerebras PyTorch package and learn how to write your own training loop for your PyTorch model by using the Cerebras PyTorch API. This package includes functions and data structures specifically tailored to facilitate the utilization of Cerebras Wafer-Scale cluster. By mastering this package, you'll be well-equipped to craft your own training loop for your PyTorch models.

Tutorials

Get accustomed with using Cerebras's Wafer-Scale to preprocess data, load data, train, and fine-tune deep neural networks using our step-by-step beginner-level instructional tutorial:

• Train and fine-tune a Large Language Model (LLM)

How-to Guides

Take advantage of our advanced how-to guides that use the Cerebras PyTorch API and other Cerebras-enhanced features:

• Train a GPT model using Maximum Update Parametrization
• Train a model with a large or small context window
• Instruction fine-tune an LLM
• Train a model with weight sparsity
• Restart a dataloader
• Port a trained and fine-tuned model to Hugging Face
• Port a Hugging Face model to Cerebras Model Zoo
• Control numerical precision level
• Dynamic Loss Scaling
• Run a Cerebras Model Zoo on GPU

Port your model

If your model is one of the supported models in Cerebras Model Zoo, you will port your model by:

• Preprocessing your data using Cerebras Model Zoo, including Hugging Face datasets
• Creating your own data processors
• Converting configurations and checkpoints from Hugging Face

If you are developing a new model, you will use the layers supported in the Cerebras Layer API and one of these approaches:

• Port your model using run function from Cerebras Model Zoo
• Create your own run function using Cerebras PyTorch API.

Cerebras Features

When developing your model:

• Understand kernel autogeneration with Autogen
• Define environment variables for input workers
• Import user-specific dependencies to Cerebras environments
• Follow best practices for CV dataloaders

When training and evaluating your model:

• Evaluate your model during training
• Summarize scalars and tensors
• Use dynamic loss scaling
• Run autoregressive inference on a batch of prompts

To boost the performance of your model:

• Measure throughput of your model
• Train with number of tokens loss scaling
• Optimizing performance with Automatic Microbatching

 

previous

Explore Cerebras Documentation

next

Cerebras Wafer-Scale Cluster